Professional Documents
Culture Documents
Ebook Machine Learning and Iot For Intelligent Systems and Smart Applications 1St Edition Online PDF All Chapter
Ebook Machine Learning and Iot For Intelligent Systems and Smart Applications 1St Edition Online PDF All Chapter
https://ebookmeta.com/product/machine-learning-techniques-and-
analytics-for-cloud-security-advances-in-learning-analytics-for-
intelligent-cloud-iot-systems-1st-edition-chakraborty/
https://ebookmeta.com/product/iot-enabled-smart-healthcare-
systems-services-and-applications-1st-edition-various-autors/
https://ebookmeta.com/product/deep-learning-and-iot-in-
healthcare-systems-paradigms-and-applications-1st-edition/
https://ebookmeta.com/product/ai-edge-and-iot-based-smart-
agriculture-intelligent-data-centric-systems-1st-edition-ajith-
abraham-editor/
https://ebookmeta.com/product/iot-machine-learning-and-
blockchain-technologies-for-renewable-energy-and-modern-hybrid-
power-systems-1st-edition-c-sharmeela/
https://ebookmeta.com/product/computer-vision-and-machine-
learning-in-agriculture-algorithms-for-intelligent-systems-
mohammad-shorif-uddin-editor/
https://ebookmeta.com/product/machine-learning-for-intelligent-
multimedia-analytics-techniques-and-applications-pardeep-kumar-
amit-kumar-singh-eds/
https://ebookmeta.com/product/design-of-intelligent-applications-
using-machine-learning-and-deep-learning-techniques-1st-edition-
ramchandra-sharad-mangrulkar-editor/
MACHINE LEARNING
AND IOT FOR
INTELLIGENT SYSTEMS
AND SMART
APPLICATIONS
Computational Intelligence in Engineering Problem
Solving
Series Editor: Nilanjan Dey
Machine Learning and IoT for Intelligent Systems and Smart Applications
Madhumathy P, M Vinoth Kumar and R. Umamaheswari
Edited by
Madhumathy P., M. Vinoth Kumar, and
R. Umamaheswari
First edition published 2022
by CRC Press
6000 Broken Sound Parkway NW, Suite 300, Boca Raton, FL 33487-2742
and by CRC Press
2 Park Square, Milton Park, Abingdon, Oxon, OX14 4RN
© 2022 selection and editorial matter, Madhumathy P, M Vinoth Kumar and
R. Umamaheswari; individual chapters, the contributors
CRC Press is an imprint of Taylor & Francis Group, LLC
Reasonable efforts have been made to publish reliable data and information, but the author
and publisher cannot assume responsibility for the validity of all materials or the
consequences of their use. The authors and publishers have attempted to trace the copyright
holders of all material reproduced in this publication and apologize to copyright holders
if permission to publish in this form has not been obtained. If any copyright material has not
been acknowledged please write and let us know so we may rectify in any future reprint.
Except as permitted under U.S. Copyright Law, no part of this book may be reprinted,
reproduced, transmitted, or utilized in any form by any electronic, mechanical, or other
means, now known or hereafter invented, including photocopying, microfilming, and
recording, or in any information storage or retrieval system, without written permission
from the publishers.
For permission to photocopy or use material electronically from this work, access
www.copyright.com or contact the Copyright Clearance Center, Inc. (CCC), 222 Rosewood
Drive, Danvers, MA 01923, 978-750-8400. For works that are not available on CCC please
contact mpkbookspermissions@tandf.co.uk
Trademark notice: Product or corporate names may be trademarks or registered trademarks
and are used only for identification and explanation without intent to infringe.
Library of Congress Cataloging-in-Publication Data
A catalog record has been requested for this book
DOI: 10.1201/9781003194415
Typeset in Times
by MPS Limited, Dehradun
Contents
Preface......................................................................................................................vii
Editors’ Biographies .................................................................................................ix
Contributors ..............................................................................................................xi
v
vi Contents
Index......................................................................................................................225
Preface
This book includes all associated topics to Machine learning, Big data, Internet of
things and applications in Internet of things. The main objective of this book is to
fetch numerous innovative studies in Machine learning, Big data and Internet of
Things. It supports the researchers, engineers and students in several interdisciplinary
domains to support realistic applications. The book presents an overview of the
different algorithms by focusing on the advantages, disadvantages and applications of
each algorithm in the field of Machine learning and IOT. The book provides machine
learning (ML) techniques to address both intelligence and configurability to various
IoT devices. The book also reports the challenges and the future directions in the IoT
and machine learning. This book comes with an energy-efficient cross layer model
and energy-related routing metric combination to prolong the lifetime of low power
IoT networks. This book deals with Machine Learning which is subset of AI that uses
computational statistics to find a mathematical model describing Input and Output
Data. Machine Learning techniques have been successfully involved in a various
applications including assistance in medical diagnosis and analyzing disease based on
clinical and laboratory symptoms with appropriate data to give more efficient result
for diagnosing disease.
Though these new skills are prodigious, they result in numerous challenges
including resource constraints of IoT devices, poor interoperability, heterogeneity of
IoT system and several privacy and security vulnerabilities. They also expose severe
IoT security challenges. Further, traditional security approaches against the most
prominent attacks are insufficient. Therefore, enabling the IoT devices to learn and
adapt to various threats dynamically and addressing them proactively need immediate
attention. In this regard, machine learning (ML) techniques are employed to address
both intelligence and reconfigurability to various IoT devices.
Outcome:
vii
Editors’ Biographies
Dr. Madhumathy P. is working as a professor at Dayananda
Sagar Academy of Technology and Management, Bengaluru,
Karnataka, India. She completed her engineering from Anna
University in 2006. M.E (gold medalist) from AVIT in 2009 and
Ph.D. from Anna University in 2015. With rich experience in
teaching for about 14 years, her area of interests include
Computer Networks, Wireless Communication, Wireless
sensor Networks, Internet of Things, Wireless Channel
Modeling, Mobile Communication and topics related to
Networks and Wireless Communication domains. Having published more than
75 papers in international, national journals and conferences, she is a life member in
ISTE and senior member from IEEE, and is serving as a reviewer for IEEE, IET,
Springer, Inderscience and Elsevier journals. She has registered and published three
Indian patent. Having received a grant for the title “A Complex Programmable Logic
Device Based Green House Monitoring System for Agriculture” from VGST, Govt. of
Karnataka, under SMYSR program, she has published a book titled “Computer
Communication Networks” with ISBN number 978-81-937245-1-4. She has acted as
publication chair for international IEEE conference held at DSATM, and has conducted
and coordinated many workshops and FDPs.
ix
x Editors’ Biographies
CONTENTS
1.1 Introduction....................................................................................................... 2
1.2 Feature Extraction ............................................................................................ 4
1.2.1 Fourier Transform (FT)........................................................................ 4
1.2.2 Short Time Fourier Transform (STFT) ...............................................4
1.2.3 Wavelet Transform............................................................................... 5
1.2.3.1 Discrete Wavelet Transform ................................................. 6
1.2.3.2 Discrete Curvelet Transform................................................. 7
1.2.3.3 Discrete Contourlet Transform.............................................. 7
1.2.3.4 Discrete Shearlet Transform.................................................. 9
1.2.3.5 Bendlet Transform...............................................................11
1.3 Classification................................................................................................... 11
1.3.1 Logistic Regression ............................................................................12
1.3.2 K-Nearest Neighbor ...........................................................................12
DOI: 10.1201/9781003194415-1 1
2 Machine Learning and IoT
1.1 INTRODUCTION
The incidence of malignant melanoma is in the majority cases fatal and increasing
worldwide. According to the 2020 Melanoma Skin Cancer Report of the Global
Cancer Observatory, there were 287,723 cases of melanoma and 1,042,056 of non-
melanoma cancers recorded globally with a greater number of cases in Australia
and the United States of America than anywhere else in the world [1]. Gender-wise,
men are 10% more likely to develop melanoma skin cancer than women and 4%
more likely to die from melanoma than women. (Figure 1.1) gives a snapshot of
growth of skin cancer from current to future projected levelsand (Table 1.1) gives
projected levels of new cases of skin cancer.
The two major types of skin cancers are Melanoma and Non-Melanoma.
Melanoma arises from malignant melanocytic cells of the epidermis and this cell
produce melanin pigment which decides the color of the skin. The malignant
melanocytic cells grow abnormally and invade other skin cells forming a big mass
of cells called the tumor. Early detection and treatment profoundly lead to prognosis
of the disease. With many skin imaging techniques developed in recent years for
assisting dermatologists to detect the melanoma, some of the techniques are (i)
Total Body Photograph [2] (ii) Ultrasonography [3] (iii) Epiluminescence
Microscopy [4] (iv) Cross Polarization Epiluminescence and (v) Optical Coherence
Tomography. Dermatologists use noninvasive methods [5] like dermoscopic and
macroscopic methods to identify melanoma in dermoscopic images. Dermoscopy
provides better diagnosis as compared to that with the naked eye [6,7].
Based on the deep residual network, the Sultana et al. [8]-developed skin cancer
classification system by using the regularized fisher platform and the convolutional
neural network to extract the low discriminative function to classify the melanoma.
Sonia et al. [9] proposed the non-subsampled contourlet transform to extract fea
tures using local binary pattern and gray level co-occurrence matrix, but due to high
dimensionality, it cannot extract texture features effectively. Many approaches have
been developed to identify melanoma from normal images. Zaquot et al. [10] used
ABCD techniques to identify melanoma performing median filtering in pre
processing and feature extraction by entropy and bifold methods. Ma et al. [11]
developed skin cancer classification system by extracting ABCDE features that
contain both shape and color features. Machine learning algorithms like decision
tree, K-nearest neighbors, and artificial neural networks were used by [12] to
compare the performance of each algorithm for melanoma classification. Also,
melanoma is identified by its seven characteristics using the so-called seven-point
checklist [13] and its three characteristics using the so-called the three-point
checklist [14].
Classifications for Melanoma Detection 3
TABLE 1.1
Global Cancer Observatory Report 2020
Estimated New Cases in 2025 340,271
Estimated New Cases in 2040 466,914
Nasir et al. [15], who extracted three features namely color, texture and
histogram-oriented gradients features for melanoma detection, selected the superior
features as input to SVM classifier for the classification of melanoma images. Rahul
et al. [16], who discussed a skin cancer classification system in which the pre
processing step is carried out by median filtering and enhancement using histogram
equalization methods, used the residual convolutional neural network as classifier to
classify the images; but with the drawback that the noise gets enhanced due to
enhancement in the preprocessing stage. Wavelet based skin cancer classification
system developed by [17] decomposes the image to give approximate and detailed
coefficients which are ploughed back into the probabilistic neural network for
further classification. Another Wavelet based resnet model was used by [18] but it
requires large amounts of training data to perform better classification than
expectation.
Donoho et al. [19] developed continuous curvelet transform for detecting curves
in the medical image but the construction of the waveform was not built-in to work
with a discrete domain directly. While deep neural network-based classification was
presented by [20] for the classification of melanoma and non-melanoma images, the
drawback of this method is that only biopsy proven images were considered in
classification module. Deep learning methods were presented by Goyal et al. [21]
for lesion segmentation in skin cancers, where the lesion are segmented exactly
from the dermoscopic images and various metrics such as specificity, sensitivity and
accuracy calculated to measure the performance of the method.
4 Machine Learning and IoT
F{f(t)} = f (t ) e j t dt (1.1)
F 1{f(t)} = f (t ) e j t dt (1.2)
j2 ki + jl
M 1 N 1
F(k, l) = i =0 j =0 f (i , j ) e M N (1.3)
1 j2 ki + jl
M 1 N 1
f(i, j) = i =0 j =0
F (K , l ) e M N (1.4)
MN
Drawbacks
The Fourier Transform does not provide time information when the signal occurs
but gives the frequency information that exists in the signal.
It is not suitable for non-stationary signals.
STFT(t , f) = x (t ) w (t t )e j t dt (1.5)
Classifications for Melanoma Detection 5
Drawbacks
• Once window size is fixed, it cannot be changed.
• STFT cannot extract texture features from images.
1 t b
a,b (t) = a, b R (1.6)
a a
Here,
a. Scale (Dilation)
b. Translation (Position)
Generally, wavelets are characterized at scale and location. Wavelets are broadly
classified as
1 t b
CWT(a, b) = <f, a, b > = dt (1.7)
a a
6 Machine Learning and IoT
1 m 1 n 1
W (j0 , m, n) x =0 y =0
f (x , y ) j0 , m, n (x, y ) (1.8)
mn
Detailed Coefficients
1 m 1 n 1
W i (j0 , m, n) = x =0 y =0
f (x , y ) j0i , m, n (x, y ) (1.9)
mn
Where, i= {H, V, D}
Drawbacks
• Wavelet transform can capture frequency content only in limited directions,
so it has poor directionality.
• Inefficient for capturing anisotropic features like lines and curves.
Width = Length2
Where,θ = Orientation
k1, k2 = Spatial Location of curvelets
f(x,y) = Input image having dimension M×N
Image decomposition in curvelet transform consists of following steps. In the
first step, the image is decomposed into sub bands by applying wavelet transform.
Then each sub band is smoothly windowed as squares by parabolic scaling and each
square is applied with discrete ridgelet transform. The wedges of parabolic shape in
the fourier plane are obtained by a radial window function and an angular window
function. The radial window is responsible for image decomposition into scales, and
the angular window is responsible for orientation and direction. Finally, curvelet
coefficients are obtained by applying IFFT to all the wedges. (Figure 1.3) depicts
curvelet transform as a combination of parabolic scaling of a sub-band into a
smooth square window, and discrete ridgelet transform applied on each square
thereafter.
Drawbacks
• Construction is easy in continuous domain and difficult to implement in
discrete domain.
• Difficult to sample on the rectangular grid.
• Curvelet involves rotations and these operators will not preserve digital
lattice.
Drawbacks
• Contourlet transform suffers from Pseudo Gibbs effect.
• It cannot detect the non-smooth corner points.
• It is not a shift invariant.
Where,
Tt is the translation operator,
DAa is the dilation, and
DSs is the shearing operator.
The dilation is given by scaling matrix and the shearing matrix
a 0
Aa = 1 wherea > 0 (1.12)
0 a2
Shearlet transform uses anisotropic dilation and orientation to precisely capture the
geometric edges. The continuous shearlet transform is represented as
3
a, s, t (x ) =a 4 (A 1B 1 (x t )) (1.14)
Figure 1.6 shows the frequency tiling containing horizontal part of the cone and
vertical part of the cone. The directional components are obtained by the translating
the shear matrix.
Drawbacks
• Shearlets cannot classify the curvature precisely.
• Shearlets are redundant in nature.
When
α = 1 ➜ Isotropic Scaling
α = 0.5 ➜ Parabolic Scaling
α = 0 ➜ Pure Directional Scaling
1.3 CLASSIFICATION
Classification is an especially important phase in medical image analysis which
predicts or classifies the unlabeled data to a set of known labeled data. Classification
generally contains two phases: namely, the training and the testing phase. In the
training phase, training datasets are used to build up a model whereas in the testing
phase, validation of model takes place to qualify performance. Some important
classifiers such as logistic regression, decision trees, K-nearest neighbors, and
FIGURE 1.7 Curve Detection by (a) Wavelet – Many Coefficients Are Needed to Capture
the Curve (b) Curvelet/Shearlet – Few Coefficients Are Needed to Capture the Curve (c)
Bendlet – Very Few Coefficients Are Needed to Capture the Curve.
12 Machine Learning and IoT
support vector machines are discussed which would be useful to classify melanoma
from dermoscopic images into normal and abnormal images.
Advantages
• Easy to implement and quite easy to train.
• Provides good accuracy when dataset is linearly separable.
• Works fast in classifying unknown records.
Disadvantages
• It cannot solve non-linear problems.
• Predicts only discrete functions.
Advantages
• Simple and intuitive.
• No training period as it is a lazy learner.
Disadvantages
• Does not work with high dimensionality.
• Large samples are needed for accuracy.
Classifications for Melanoma Detection 13
n i i
Entropy = i=1 p log2 p (1.17)
t t
2
n i
Gini = 1 i =1
p (1.18)
t
Advantages
• Nonlinear relationship parameters will not affect performance.
• Easy to interpret.
• Need of domain knowledge is not necessary.
Disadvantages
• Always restricted to one output attribute.
• Unstable classifier depends on the dataset.
• It generates a complex tree when the dataset is numeric.
It consists of the separating plane that separates data into two class or multiclass.
It also contains two margin lines that run parallel with hyperplane and maintains
some distance to classify the data effectively. The margin lines are separated by
maximum distance and each margin line passes through one of the vectors in po
sitive and negative class. These vectors are called support vectors. Linear separable
SVM can separate the data points easily by constructing a hyperplane. The data
points cannot be easily separated in case of nonlinear separable SVM. (Figure 1.9)
depicts SVM hyperplane for (A) linear data to be separated and (B) non-linear data.
In Non-linear case, it converts low dimension data points to high dimension
datapoints using kernel function. The three types of kernel function are given
below. (Figure 1.10) depicts a non-linear SVM classification using Radial Basis
Function for its kernel function.
Advantages
• Provides better accuracy.
FIGURE 1.9 (a) Linear Separable Data (b) Non linear Separable Datad SP Learned
Dictionaries with CS.
Disadvantages
• Selection of right kernels.
• Training the data takes more time.
• Computationally expensive.
abnormal. (Figure 1.12) depicts the block diagram of the algorithm for classifying
images in the SCC system.
The performance of the system is measured by sensitivity, specificity, and ac
curacy. Table 1.2 below gives the accuracy obtained from the set of 75 and 100
coefficients of shearlet transform and K-NN classifier.
It is observed from Table 1.2 that the second level of shearlet features made from
100-point coefficients yield better performance compared to other levels of shearlet
decomposition. The maximum accuracy of 96% is obtained in the second level of
decomposition using shearlet transform and K-NN classifier. (Figure 1.13) depicts
the K-NN classifier performance for 75 coefficients vis-a-vis 100 coefficients for
four levels of decomposition.
TABLE 1.2
Performance of K-NN Classifier
Shearlet Transform Decomposition Level K-NN Classification Accuracy (%)
75 Coefficients 100 Coefficients
FIGURE 1.14 Block Diagram of Bendlet and SVM Based SCC System.
TABLE 1.3
Performance Metrics of SCC System
Performance Description Formula
Measure
1.5 CONCLUSION
This study provides a brief description on various image representation systems
and classification techniques used to classify dermoscopic images into normal and
abnormal images. In work I, shearlet transform and K-NN classifier gives maximum
accuracy of 96% in the second level of shearlet decomposition with 100 coeffi
cients. In work II, bendlet transform for feature extraction and SVM for classifi
cation give maximum accuracy of 98.3% in level 3 decomposition and direction 8.
Finally, all the image representation systems like wavelet, curvelet, contourlet,
shearlet and bendlet are compared and performance metrics like specificity,
Classifications for Melanoma Detection 19
TABLE 1.4
Classifier Performance of Normal and Abnormal Phase
Level of Number of Performance Measures
Decomposition Directions Tr Po Fa Ne Tr Ne Fa Po Sn (%) Sp (%) Ac (%)
sensitivity and accuracy are measured. From the results, we come to the conclusion
that bendlet provides better accuracy compared to all other image representation
systems. Wavelet is the worst performer whereas contourlet is the better performer
than either curvelet or wavelet.
REFERENCES
[1]. del Marmol V, lipira K, ‘2020 Skin cancer report- global cancer observatory’, 2020.
[2]. Dengel LT, Petroni GR, Judge J, Chen D, Acton ST, Schroen AT, Slingluff CL,
‘Total body photography for skin cancer screening’, International Journal of
Dermatology, 54(11): ISSN:1365-4632, 1250–1254, 2015, Wiley.
[3]. Srivastava A, Hughes B, Hughes L, Woodcock J, ‘Doppler ultrasound as an adjunct
to the differential diagnosis of pigmented skin lesions’, British Journal of Surgery,
73: 790–792, ISSN: 1365-2168, 1986, John Wiley & Sons.
[4]. Pehamberger H, Binder M, Steiner A, Wolff K, ‘In vivo epiluminescence micro
scopy: improvement of early diagnosis of melanoma’, Journal of Investigative
Dermatology, 100(3), ISSN: 0022-202X1993, Elsevier.
[5]. Banerjee I ‘Brain tumor image segmentation and classification using SVM, CLAHE
AND ARKFCM’, Intelligent decision support systems, applications in signal pro
cessing, 978-3-11-062110-5, 53–70, October, 2019
[6]. Banerjee I ‘IOT based fluid and heartbeat monitoring for advanced health care’,
Classification techniques for medical image analysis and computer aided diagnosis
volume 4 , ISBN: 978-0-12-818004-4, 2019 .
[7]. Madhumathy P, Prasanth R, Srilekha ‘Eye movement detection for paralyzed patient
using pressure sensor’, International Journal of Scientific & Engineering Research,
7(11): 395–397, November, 2016 ISSN 2229-5518.
[8]. Sultana NN, Manda, B, Puhan NB, ‘Deep residual network with regularised fisher
framework for detection of melanoma’, IET Computer Vision, 12(8): 1096–1104,
ISSN 1751-8644, 2018, IET Digital Library.
[9]. Sonia R et al., ‘Melanoma image classification system by NSCT features and Bayes
classification’, International Journal of Advances in Signal and Image Sciences,
2(2): 27–33, -ISSN: 2457-0370, 2016.
[10]. Zaqout I, ‘Diagnosis of skin lesions based on dermoscopic images using image
processing techniques’, International Journal of Signal Processing, Image
Processing and Pattern Recognition, 9(9): 189–204, ISSN: 20054254, 2207970X,
2016, Science and Engineering Support Society.
[11]. Ma Z, Tavares JM, ‘Effective features to classify skin lesions in dermoscopic
images’, Expert Systems with Applications, 84: 92–101, ISSN: 0957-4174, 2017,
Elsevier.
[12]. Ozkan IA, Koklu M, ‘Skin lesion classification using machine learning algorithms’,
International Journal of Intelligent Systems and Applications in Engineering, 5(4),
ISSN:2147-6799 285-289, 2017, Academic Publisher Science.
[13]. Argenziano G, Catricalà C, Ardigo, et al. ‘Seven-point checklist of dermoscopy
revisited’, British Journal of Dermatology, 164(4): 785–790, ISSN: 1365-2133,
2011, Wiley.
[14]. Zalaudek I, Argenziano G, Soyer, H, et al. ‘Three-point checklist of dermoscopy: an
open internet study’, British Journal of Dermatology, 154(3): 431–437, ISSN: 1365-
2133, 2006, Wiley.
[15]. Nasir M, Attique Khan M, Sharif M, Lali IU, Saba T, Iqbal T, ‘An improved
strategy for skin lesion detection and classification using uniform segmentation and
Classifications for Melanoma Detection 21
CONTENTS
2.1 Introduction..................................................................................................... 23
2.2 Machine Learning in MSA Design ...............................................................25
2.3 Application of MSA in IOT ..........................................................................26
2.4 Design & Analysis of MSA Using ANN......................................................27
2.4.1 Artificial Neural Network ..................................................................27
2.5 Results and Discussion...................................................................................29
2.6 Design of Microstrip Antenna and Characterization Using SVM Method..30
2.7 Design of MSA for IoT Applications ...........................................................31
2.8 Conclusion ......................................................................................................32
References................................................................................................................32
2.1 INTRODUCTION
The essential advantage in using microstrip antennas lies in the fact that they possess high
power gain and omni-directional radiation patterns that are particularly useful in appli
cations such as mobile communications, given mobile services need wideband frequency
operations and will find these antennas suitable for attaining higher accuracy in their
services offering. It is a critical design issue to implement printed antennas for achieving
multiple application level services provided by mobile communication applications.
DOI: 10.1201/9781003194415-2 23
24 Machine Learning and IoT
Accuracy is also a major factor to be considered in such design cases. There are multiple
parameters that decide the performance of the antenna such as patch dimensions of mi
crostrip antenna. Fabrication technique used in microstrip antenna design is easy and low
cost [1,2]. For example, photo etching mechanism is another approach of fabrication of
microstrip antennas which is commonly termed as patch, where various patterns and
shapes for patches like square, hexagonal, rectangular, trapezoidal etc. are commonly in
use. In this chapter, we have proposed an artificial neural network (ANN) modeling
method for synthesis and analysis of microstrip antennas. Thus, an antenna is a radiating
device that radiates electromagnetic energy in desired directions, where omni-directional;
semi-directional and directional antennas are used. Antennas are used to redirect the radio
frequencies provided by the transmitter to the receiver in free space [3]. Antenna is re
quired to cover various frequency bandwidths or wide frequency band. It is expected that
the antenna size should be small and its performance can be affected by changing the
geometries of mounting devices [4]. Nowadays, demand for low cost antenna can be
fulfilled by reduced size and compressed architecture for antennas. A microstrip antenna
has several benefits when compared to supplementary antennas such as low weight, low
cost, low profile, low scattering cross section, option of undeviating and rounded polar
ization with particular feed and can be simply incorporated for microwave circuits [5].
There are numerous procedures for investigation of microstrip antenna with some of the
popular procedures being transmission line, full wave and cavity. The earliest model of
microstrip antenna, among all, is the transmission line model because of its good physical
insight, where the design procedure for this antenna assumed this information with di
electric constant of the substrate (ε), resonant frequency (f) and substrate height (h) [6]. A
microstrip antenna is well-defined according to an array of dual radiating contracted
apertures, with height h, width w and implanted L distance apart. The antenna design
procedure is as follows: To find the width of the patch in practical cases for better radiation
efficiencies, we follow Equation (2.1).
1 2
w= (2.1)
2f o 0 +1
+1 1 1
reff = + (2.2)
2 2 1 + 12 w
h
Patch width and length are calculated by Equations (2.3) and (2.4).
v0 2
w = (2.3)
2fr r + 1
Machine Learning Based Antenna Design 25
v0
l = = 2 L (2.4)
2fr reff
L
= 0.412
( reff + 0.3) h + 0.264 ( w
) (2.5)
h w
( reff 0.258) h + 0.8 ( )
Here L is the length expansion in presence fringing effects. Equation (2.6) finds
the effective dielectric constant.
1/2
r +1 r 1 h
reff = + 1 + 12 (2.6)
2 2 w
Antennas have been designed for a given application according to the required
performances. Nowadays, the task is to catch the available symmetrical factors of
the patch (like dimensions of patch, dimensions of ground and feed position) that
consume a portion of time due to trial and error practice. To decrease this con
sumption of time, vast amount of techniques are used (such as optimum algorithms,
ANN techniques etc).
yi = f ( wij xj ) (2.7)
yi = f ( wij xj ) (2.9)
28 Machine Learning and IoT
In Figure 2.1 the network consists of 3 input nodes, 7 hidden nodes and 2 output
nodes. Sigmoid function is used for activation and Bias input is considered as 1 for
both layers.
Here the learning rate is considered in a range of 0 and 1 with random weights.
Following are the steps followed in the proposed work:
1. Connecting weights updating is done between of the output layer and the
hidden layer neurons.
2. Also update connecting weights between input layer and the hidden layer
neurons.
Step 3: Keep repeating step 1 and step 2 to update weights at occurrence of each
event (Tables 2.1 and 2.2).
Step 4: Terminate the process when negligibly small value of the error is ob
tained (Figures 2.2 and 2.3).
TABLE 2.1
Measurement of the proposed antenna
Substrate Size 18 mm × 20 mm
Substrate Height 2 mm
Microstrip Feed Line length 2 mm
Microstrip Feed Line width 0.5 mm
Patch length 14 mm
`Patch width 12 mm
TABLE 2.2
Comparson of results of ANN output W,L with rest to r, h , fr
r h fr W(Simulated) L(Simulated) W(ANN) L(ANN)
and testing the network, past data sets are used. Theoretical results are matched with
measured results with maximum extent. Error obtained is infinitesimally small
(Figures 2.4 and 2.5).
1863 július.
I.